在上一篇文章OpenGLES + MediaCodec 短视频分段录制实现与无丢帧录制优化中说到了MediaCodec录制视频的一些优化思路。经过上一步,我们就实现了短视频的分段录制功能以及录制视频帧率的优化。在得到多段视频之后,我们如何将这些细小的视频文件合成一个视频文件呢? 这是有好多种方法,比如通过MP4Parser、FFmpeg等进行合成。这里介绍如何使用Android原生的MediaExtractor + MediaMuxer进行对多段MP4视频文件进行合成。
1、MediaExtractor 提取媒体信息
MediaExtractor 主要的作用是从数据源中提取媒体数据,并将媒体数据传递给解复用器(通常是解码器)
使用MediaExtractor 一般是这样的,不断循环读取VideoPath/AudioPath,将媒体信息从文件中读取出来, 筛选出不同的轨道trackIndex并通过getTrackFormat(trackIndex)方法获取对应的轨道信息:
// MediaExtractor拿到多媒体信息,用于MediaMuxer创建文件
while (videoIterator.hasNext()) {
String videoPath = (String) videoIterator.next();
MediaExtractor extractor = new MediaExtractor();
try {
extractor.setDataSource(videoPath);
} catch (Exception ex) {
ex.printStackTrace();
}
int trackIndex;
if (!hasVideoFormat) {
trackIndex = selectTrack(extractor, "video/");
if(trackIndex < 0) {
Log.e(TAG, "No video track found in " + videoPath);
} else {
extractor.selectTrack(trackIndex);
mVideoFormat = extractor.getTrackFormat(trackIndex);
hasVideoFormat = true;
}
}
if (!hasAudioFormat) {
trackIndex = selectTrack(extractor, "audio/");
if(trackIndex < 0) {
Log.e(TAG, "No audio track found in " + videoPath);
} else {
extractor.selectTrack(trackIndex);
mAudioFormat = extractor.getTrackFormat(trackIndex);
hasAudioFormat = true;
}
}
extractor.release();
if (hasVideoFormat && hasAudioFormat) {
break;
}
}
/**
* 选择轨道
* @param extractor MediaExtractor
* @param mimePrefix 音轨还是视轨
* @return
*/
private int selectTrack(MediaExtractor extractor, String mimePrefix) {
// 获取轨道总数
int numTracks = extractor.getTrackCount();
// 遍历查找包含mimePrefix的轨道
for(int i = 0; i < numTracks; ++i) {
MediaFormat format = extractor.getTrackFormat(i);
String mime = format.getString("mime");
if (mime.startsWith(mimePrefix)) {
return i;
}
}
return -1;
}
2、创建MediaMuxer并添加相应的轨道
通过MediaExtractor 拿到媒体的轨道trackIndex 和 轨道信息trackFormat之后,我们就可以创建一个MediaMuxer复用器并添加相应的轨道(音频轨道、视频轨道等):
try {
mMuxer = new MediaMuxer(mDestPath, MediaMuxer.OutputFormat.MUXER_OUTPUT_MPEG_4);
} catch (IOException e) {
e.printStackTrace();
}
if (hasVideoFormat) {
mOutVideoTrackIndex = mMuxer.addTrack(mVideoFormat);
}
if (hasAudioFormat) {
mOutAudioTrackIndex = mMuxer.addTrack(mAudioFormat);
}
mMuxer.start();
3、合并视频
根据前面得到的媒体信息、媒体轨道等信息就可以进行合并了。合并的流程大体如下所示:
① 选择视频的轨道(音频轨道、视频轨道等),判断是否存在音视频轨道,并获取对应轨道的index。
② 判断视频的视频轨道和视频轨道是否存在,如果都不存在,则该视频文件很可能损坏了,合并失败,并退出。
③ 如果存在音频轨道、视频轨道,进入合并阶段
④ 通过MediaExtractor的readSampleData()方法从视频中读取音频轨道、视频轨道的数据,并判断读取出来的数据帧是否大于0,如果大于0,则用getSampleTime()方法读取数据帧的pts,并通过getSampleFlags()读取flag标志,然后创建一个BufferInfo类用来保存这些帧数据。
⑤ 将MediaExtractor提取的数据,通过MediaMuxer 的writeSampleData()方法将媒体数据写入到新的文件
⑥ 遍历下一个视频文件,并计算从下一个视频文件提取出来的媒体数据的PTS,用于下一次的视频合成
⑦ 释放MediaMuxer、MediaExtractor
至此,我们就把多段MP4视频合成功能的流程介绍完了,完整的合并代码如下所示:
@TargetApi(18)
public class VideoCombiner {
private static final String TAG = "VideoCombiner";
private static final boolean VERBOSE = true;
// 最大缓冲区(1024 x 1024 = 1048576、1920 x 1080 = 2073600)
// 由于没有录制1080P的视频,因此用1024的Buffer来缓存
private static final int MAX_BUFF_SIZE = 1048576;
private List mVideoList;
private String mDestPath;
private MediaMuxer mMuxer;
private ByteBuffer mReadBuf;
private int mOutAudioTrackIndex;
private int mOutVideoTrackIndex;
private MediaFormat mAudioFormat;
private MediaFormat mVideoFormat;
private VideoCombineListener mCombineListener;
public interface VideoCombineListener {
/**
* 合并开始
*/
void onCombineStart();
/**
* 合并过程
* @param current 当前合并的视频
* @param sum 合并视频总数
*/
void onCombineProcessing(int current, int sum);
/**
* 合并结束
* @param success 是否合并成功
*/
void onCombineFinished(boolean success);
}
public VideoCombiner(List videoList, String destPath, VideoCombineListener listener) {
mVideoList = videoList;
mDestPath = destPath;
mCombineListener = listener;
mReadBuf = ByteBuffer.allocate(MAX_BUFF_SIZE);
}
/**
* 合并视频
* @return
*/
@SuppressLint("WrongConstant")
public void combineVideo() {
boolean hasAudioFormat = false;
boolean hasVideoFormat = false;
Iterator videoIterator = mVideoList.iterator();
// 开始合并
if (mCombineListener != null) {
mCombineListener.onCombineStart();
}
// MediaExtractor拿到多媒体信息,用于MediaMuxer创建文件
while (videoIterator.hasNext()) {
String videoPath = (String) videoIterator.next();
MediaExtractor extractor = new MediaExtractor();
try {
extractor.setDataSource(videoPath);
} catch (Exception ex) {
ex.printStackTrace();
}
int trackIndex;
if (!hasVideoFormat) {
trackIndex = selectTrack(extractor, "video/");
if(trackIndex < 0) {
Log.e(TAG, "No video track found in " + videoPath);
} else {
extractor.selectTrack(trackIndex);
mVideoFormat = extractor.getTrackFormat(trackIndex);
hasVideoFormat = true;
}
}
if (!hasAudioFormat) {
trackIndex = selectTrack(extractor, "audio/");
if(trackIndex < 0) {
Log.e(TAG, "No audio track found in " + videoPath);
} else {
extractor.selectTrack(trackIndex);
mAudioFormat = extractor.getTrackFormat(trackIndex);
hasAudioFormat = true;
}
}
extractor.release();
if (hasVideoFormat && hasAudioFormat) {
break;
}
}
try {
mMuxer = new MediaMuxer(mDestPath, MediaMuxer.OutputFormat.MUXER_OUTPUT_MPEG_4);
} catch (IOException e) {
e.printStackTrace();
}
if (hasVideoFormat) {
mOutVideoTrackIndex = mMuxer.addTrack(mVideoFormat);
}
if (hasAudioFormat) {
mOutAudioTrackIndex = mMuxer.addTrack(mAudioFormat);
}
mMuxer.start();
// MediaExtractor遍历读取帧,MediaMuxer写入帧,并记录帧信息
long ptsOffset = 0L;
Iterator trackIndex = mVideoList.iterator();
int currentVideo = 0;
boolean combineResult = true;
while (trackIndex.hasNext()) {
// 监听当前合并第几个视频
currentVideo++;
if (mCombineListener != null) {
mCombineListener.onCombineProcessing(currentVideo, mVideoList.size());
}
String videoPath = (String) trackIndex.next();
boolean hasVideo = true;
boolean hasAudio = true;
// 选择视频轨道
MediaExtractor videoExtractor = new MediaExtractor();
try {
videoExtractor.setDataSource(videoPath);
} catch (Exception e) {
e.printStackTrace();
}
int inVideoTrackIndex = selectTrack(videoExtractor, "video/");
if(inVideoTrackIndex < 0) {
hasVideo = false;
}
videoExtractor.selectTrack(inVideoTrackIndex);
// 选择音频轨道
MediaExtractor audioExtractor = new MediaExtractor();
try {
audioExtractor.setDataSource(videoPath);
} catch (Exception e) {
e.printStackTrace();
}
int inAudioTrackIndex = selectTrack(audioExtractor, "audio/");
if (inAudioTrackIndex < 0) {
hasAudio = false;
}
audioExtractor.selectTrack(inAudioTrackIndex);
// 如果存在视频轨道和音频轨道都不存在,则合并失败,文件出错
if (!hasVideo && !hasAudio) {
combineResult = false;
videoExtractor.release();
audioExtractor.release();
break;
}
boolean bMediaDone = false;
long presentationTimeUs = 0L;
long audioPts = 0L;
long videoPts = 0L;
while (!bMediaDone) {
// 判断是否存在音视频
if(!hasVideo && !hasAudio) {
break;
}
int outTrackIndex;
MediaExtractor extractor;
int currentTrackIndex;
if ((!hasVideo || audioPts - videoPts <= 50000L) && hasAudio) {
currentTrackIndex = inAudioTrackIndex;
outTrackIndex = mOutAudioTrackIndex;
extractor = audioExtractor;
} else {
currentTrackIndex = inVideoTrackIndex;
outTrackIndex = mOutVideoTrackIndex;
extractor = videoExtractor;
}
if (VERBOSE) {
Log.d(TAG, "currentTrackIndex: " + currentTrackIndex
+ ", outTrackIndex: " + outTrackIndex);
}
mReadBuf.rewind();
// 读取数据帧
int frameSize = extractor.readSampleData(mReadBuf, 0);
if (frameSize < 0) {
if (currentTrackIndex == inVideoTrackIndex) {
hasVideo = false;
} else if (currentTrackIndex == inAudioTrackIndex) {
hasAudio = false;
}
} else {
if (extractor.getSampleTrackIndex() != currentTrackIndex) {
Log.e(TAG, "got sample from track "
+ extractor.getSampleTrackIndex()
+ ", expected " + currentTrackIndex);
}
// 读取帧的pts
presentationTimeUs = extractor.getSampleTime();
if (currentTrackIndex == inVideoTrackIndex) {
videoPts = presentationTimeUs;
} else {
audioPts = presentationTimeUs;
}
// 帧信息
BufferInfo info = new BufferInfo();
info.offset = 0;
info.size = frameSize;
info.presentationTimeUs = ptsOffset + presentationTimeUs;
if ((extractor.getSampleFlags() & MediaCodec.BUFFER_FLAG_KEY_FRAME) != 0) {
info.flags = MediaCodec.BUFFER_FLAG_KEY_FRAME;
}
mReadBuf.rewind();
if (VERBOSE) {
Log.d(TAG, String.format("write sample track %d, size %d, pts %d flag %d",
Integer.valueOf(outTrackIndex),
Integer.valueOf(info.size),
Long.valueOf(info.presentationTimeUs),
Integer.valueOf(info.flags))
);
}
// 将读取到的数据写入文件
mMuxer.writeSampleData(outTrackIndex, mReadBuf, info);
extractor.advance();
}
}
// 当前文件最后一帧的PTS,用作下一个视频的pts
ptsOffset += videoPts > audioPts ? videoPts : audioPts;
// 当前文件最后一帧和下一帧的间隔差40ms,默认录制25fps的视频,帧间隔时间就是40ms
// 但由于使用MediaCodec录制完之后,后面又写入了一个OES的帧,导致前面解析的时候会有时间差
// 这里设置10ms效果比40ms的要好些。
ptsOffset += 10000L;
if (VERBOSE) {
Log.d(TAG, "finish one file, ptsOffset " + ptsOffset);
}
// 释放资源
videoExtractor.release();
audioExtractor.release();
}
// 释放复用器
if (mMuxer != null) {
try {
mMuxer.stop();
mMuxer.release();
} catch (Exception e) {
Log.e(TAG, "Muxer close error. No data was written");
} finally {
mMuxer = null;
}
}
if (VERBOSE) {
Log.d(TAG, "video combine finished");
}
// 合并结束
if (mCombineListener != null) {
mCombineListener.onCombineFinished(combineResult);
}
}
/**
* 选择轨道
* @param extractor MediaExtractor
* @param mimePrefix 音轨还是视轨
* @return
*/
private int selectTrack(MediaExtractor extractor, String mimePrefix) {
// 获取轨道总数
int numTracks = extractor.getTrackCount();
// 遍历查找包含mimePrefix的轨道
for(int i = 0; i < numTracks; ++i) {
MediaFormat format = extractor.getTrackFormat(i);
String mime = format.getString("mime");
if (mime.startsWith(mimePrefix)) {
return i;
}
}
return -1;
}
}
4、如何使用
为了方便使用,我们可以创建一个Manager来管理合并视频功能,这样我们就可以在Activity、Fragment里面直接通过调用VideoCombineManager.getInstance().startVideoCombiner()方法,传递需要合并的路径列表、输出路径、合并状态监听接口参数,使用起来非常方便:
public final class VideoCombineManager {
private static final String TAG = "VideoCombineManage";
private static VideoCombineManager mInstance;
public static VideoCombineManager getInstance() {
if (mInstance == null) {
mInstance = new VideoCombineManager();
}
return mInstance;
}
/**
* 初始化媒体合并器
* @param videoPath
* @param destPath
*/
public void startVideoCombiner(final List videoPath, final String destPath,
final VideoCombiner.VideoCombineListener listener) {
new Thread(new Runnable() {
@Override
public void run() {
VideoCombiner videoCombiner = new VideoCombiner(videoPath, destPath, listener);
videoCombiner.combineVideo();
}
}).start();
}
}
5、MediaExtractor + MediaMuxer 合并视频的优缺点
在录制视频合成方面,MediaExtractor + MediaMuxer 比较适合应用内简单的视频合成场景,比如在相机里面录制了多段视频,然后跳转至预览页面直接合成视频这样的场景。如果要做短视频SDK,估计有点力不从心,比如短视频SDK里面,需要添加字幕、添加mkv等功能,MediaExtractor + MediaMuxer组合是做不到的。需要做短视频SDK的童鞋,还是老老实实用FFmpeg这样成熟的框架吧。